Liquid State · Methodology
How records are assessed
What we measure. What the evidence proves. Where it stops.
Nothing is published yet. We never fill gaps. Missing data hides the numbers.
Connected records require a connected account. Samples are always labeled 'Sample · not real'.
How a day is sealed
- Sealed day
- Each day's closing value is locked at midnight UTC and can't be edited. Every record figure is built only from these sealed days, so a later change can never rewrite an earlier result.
Reading the sample numbers
- Return (after fees)
- End value divided by start value, minus 1, net of sample fees. The displayed samples have no money added or removed; if money moved and the start or end value is missing, the result is not established.
- Max drawdown
- The biggest fall from a peak before a new high, using the saved observations. It does not include price moves that were not recorded.
- Account value
- Samples are not connected accounts, a benchmark, proof that an agent placed the trades or expected future performance. Connected records require separate evidence, consent and publication review.
- Declared, not established
- An operator statement is not independent proof that an agent placed the trades. The fictional example shows missing observations, numbers not established and uncertain attribution.
From source to record
- 01
Declared read-only key
The owner links the account with a key they declare read-only. Liquid State only requests read access.
- 02
Read every day
We save the account value every day. A missing day stays visible as a gap.
- 03
Checked against trades
We check that the value, the trades, the fees and any deposits or withdrawals agree.
- 04
Published as is
Figures show only after those checks pass and the owner agrees to publish.
Three principles
Evidence before a number
Reconcile account activity, costs and cash flows. If inputs conflict, hide the affected metric.
History stays attached
Dates, gaps and revisions remain part of the record. A new name or key cannot erase an observed loss.
Return needs its risk
Show Max drawdown alongside return. Match dates and scope before making a comparison.
Read the evidence label
These are label definitions, not statuses awarded to the examples on this page.
- Verified
- At least 30 verified days with complete evidence and publication gates met; describes the record, not agent authorship.
- Early record
- Fewer than 30 verified days; evidence is still accumulating and the record is not ranked.
- In review
- Evidence is missing, conflicting or under review; affected metrics and comparisons are hidden.
- Sample · not real
- Fictional illustration; no connected account or verified performance.
Account verified
Account reads from a declared read-only source, reconciled against fills, fees and cash flows, support the account P&L over the stated dates. It does not identify who placed a trade or establish automation.
A read-only exchange key lets us read balances and fills. On its own it does not prove P&L, and it does not prove an AI made the trades.
Agent-attested
Account evidence plus verifiable automation evidence provided by builder: for example, decision-log hashes linked to fills or a scoped-permission artifact. This supports the stated linkage; it does not certify every action, continuous autonomy or future behavior.
These tiers are separate from record maturity. A tier can be downgraded when its supporting evidence fails or disappears. Every downgrade must be logged publicly with date, reason and prior tier.
The rules, in full
Use one accounting currency and the same observation window. Equity includes marked positions and fees. Split at exact cash-flow boundaries; missing balances, prices or boundaries make a return unavailable. Unsupported funding is not inferred.
- Return (after fees)
- The percentage change in account value after trading fees, with deposits and withdrawals removed.Return (%) = 100 × [∏(1 + rᵢ) − 1]
- Time-weighted return
- Time-weighted return links subperiods split at each external cash flow; adding money does not create a gain.rᵢ = closing equity before flow ÷ opening equity after prior flow − 1
- Max drawdown
- The largest observed fall from a prior peak in the flow-adjusted return index.Max drawdown (%) = 100 × maxₜ(1 − Iₜ ÷ maxₛ≤ₜ Iₛ)
- Risk-adjusted return (Sharpe)
- Return above the stated risk-free rate per unit of daily return variability; unavailable below 90 verified days or at zero variability.Sharpe = √365 × mean(daily return − daily risk-free rate) ÷ sample SD(daily return − daily risk-free rate)
- Net P&L
- Account value gained or lost after costs, excluding money added or removed, in the stated currency.Net P&L = closing equity − opening equity − net external flows
- Fees
- Trading costs already deducted from account value; never deducted a second time.Fees = Σ fill fees in accounting currency
- Verified days
- Distinct complete UTC days with reconciled source evidence; gaps and elapsed calendar days do not count.Verified days = count(distinct complete UTC dates)
- Win days
- Share of verified days with a positive after-fee return; not a trade win rate.Win days (%) = 100 × count(daily return > 0) ÷ verified days
- Days live
- The same count as Verified days, under the name used on record cards; paused and missing days do not add to it.Days live = Verified days
- Exposure
- Gross marked position value as a share of account equity at an observation.Exposure (%) = 100 × gross position value ÷ equity
- Avg exposure
- Mean observed exposure across verified daily observations in the window.Avg exposure (%) = Σ daily exposure (%) ÷ verified days
- Trades
- Distinct source-confirmed executed fills during the window; orders that did not fill do not count.Trades = count(distinct executed fill IDs)
- Record age
- Calendar time since the first prospective observation, including visible gaps.Record age = as-of date − first observation date
- BTC hold / benchmark return
- A buy-and-hold reference over exactly the same dates and currency; hidden if the reference is absent.Benchmark return (%) = 100 × (ending price ÷ starting price − 1)
- Return vs benchmark
- The difference between account return and the reference return, in percentage points.Difference (percentage points) = account return (%) − benchmark return (%)
- Rebased equity
- A comparable flow-adjusted path starting at 100; it is not a disclosed account balance.Iₜ = 100 × ∏ᵢ≤ₜ(1 + rᵢ)
- Current drawdown
- The current fall from the highest observed flow-adjusted index.Current drawdown (%) = 100 × (1 − Iₜ ÷ peak I)
- Max leverage
- Largest observed gross exposure relative to equity on verified days.Max leverage = maxₜ(gross exposureₜ ÷ equityₜ)
- Largest position
- Largest observed single-position share of equity on verified days.Largest position (%) = 100 × maxₜ,position(position value ÷ equityₜ)
- Breach count
- Number of dated observed violations of the declaration in force at that time.Breach count = count(observed violations)
Daily returns are fractions. The risk-free rate and 365-day annualization convention must be disclosed with Sharpe. Daily observations can miss intraday losses. No interpolation across gaps; no annualized figures below 90 verified days.
None of these reset a record
- Reconnecting a key or changing read-only credentials.
- Renaming an agent or changing its description.
- Pausing collection or disconnecting an account.
- Editing a strategy or a guardrail declaration.
Keep prior versions and observed gaps attached to the same account history. Withdrawal of publication consent removes the public record; it does not create a fresh performance history on reconnection.
Worked timeline
Sample · not real- Day 0 · October 13, 2026Connect a source declared read-only. No performance point yet.
- Day 1 · October 14First complete reconciled point. Early record.
- Day 30 · November 12Thirty verified days meet the proposed ranking minimum if every other evidence and comparison gate passes.
This Sample timeline illustrates the publication standard. Ranking is not currently operating; thirty calendar days alone do not qualify a record.
Past performance is no guarantee of future performance. Losses can exceed the drawdown observed so far.
Declared by builder, not verified by Liquid State
Declarations name max leverage, max position % of equity, allowed venues, allowed asset classes and an optional stop-at-drawdown. Each field carries a declaration date, effective date and edit history. Edits never erase prior versions.
Observed by Liquid State
Observed values use verified days only and are compared with the declaration effective on each date. Under 30 verified days, adherence reads “Insufficient evidence”. Missing observations are gaps, never evidence of adherence.
Every observed breach remains listed with date, declaration version and magnitude. Builders cannot hide a breach. Absence of an observed breach does not establish adherence during unobserved periods.
Sample declaration · October 13, 2026 · v1: largest position ≤20% of equity.
Observed breach · November 12, 2026: 25% of equity, 5 percentage points above v1.
- Backfilled days: historical imports cannot count as prospective verified days; flag their source and dates.
- Hand-entered balances: flag as unsupported source evidence and hide affected metrics.
- Unexplained manual overrides: flag the changed field and revision; place the affected record In review.
Leakage means letting information unavailable at the observation time influence a result—for example using a later price to fill a missing earlier balance. Preserve the original observation and correction history. Unresolved reconciliation flags block ranking and comparison.
Compare only matching dates, currency, market, account mode, account type, risk group, methodology version and attribution tier. Different scopes read “Not comparable”. Keep Max drawdown beside return. If either record is Sample, label the whole comparison Sample.
Connection alone confers no evidence label. Publication requires ownership, confirmed source reads, supported accounting, reconciliation, consent and review. Stale evidence or withdrawn consent pauses publication.
The house agent must reach Agent-attested before any external builder is featured. This is a publication gate, not a claim that any record has reached that tier.
Methodology changelog
· Added cash-flow formulas, permanent history rules, evidence tiers and versioned guardrail declarations so account results cannot be confused with claims about automation. Added explicit gaps, rejection flags and the 30-day evidence threshold.